A fuzzy algorithm for learning vector quantization

This paper proposes a fuzzy algorithm for learning vector quantization, which can train feature maps to function as pattern classifiers through an unsupervised learning process. The development of the proposed algorithms is based on the minimization of a fuzzy objective function, formed as the weighted sum of the squared Euclidean distances between an input vector, which represents a feature vector, and the weight vectors of the map, which represent the prototypes. The distances between each input vector and the prototypes are weighted by a set of generalized membership functions, which regulate the competition between various prototypes for each input and, thus, determine the strength of the attractions between each input and the prototypes during the learning process. A specific set of generalized membership functions provide the basis for the derivation of the FALVQ 1 algorithm. The properties of the proposed algorithm are analytically studied. The efficiency of the FALVQ 1 algorithm is also evaluated by its use in vector quantizer design required for image compression